JSSC 2020第3期Digital Circuits65nm

OPTIMO: A 65-nm 279-GOPS/W 16-b Programmable Spatial-Array Processor with On-Chip Network for Solving Distributed Optimizations via the Alternating Direction Method of Multipliers

OPTIMO是一款65纳米工艺的16位可编程空间阵列处理器，用于分布式优化问题求解。

65nm CMOS, 279 GOPS/W

空间阵列处理器分布式优化ADMM算法可编程架构能效优化

▸创新点1：49核可编程空间阵列设计，采用65nm工艺实现高密度集成，每个核心支持16位精度计算，通过并行架构显著提升ADMM算法的计算效率，实测峰值能效达279 GOPS/W。

▸创新点2：分层多播网络架构，优化片上通信效率，支持动态路由和低延迟数据传输，显著减少分布式优化算法中的通信开销，提升多核协同计算性能。

▸创新点3：支持ADMM算法的六种模板算法，提供高度可编程性，适用于多种约束优化问题，通过硬件加速实现快速收敛，扩展了处理器的应用范围。

▸创新点4：采用交替方向乘子法（ADMM）的硬件实现，通过分解决策向量和并行更新策略，有效解决大规模优化问题，展示了在实时信号处理和机器学习中的潜力。

Abstract

This article presents OPTIMO, a 65-nm, 16-b, fully programmable, spatial-array processor with 49 cores and a hierarchical multi-cast network for solving distributed opti- mizations via the alternating direction method of multipli- ers (ADMM). ADMM is a projection-based method for solving generic-constrained optimizations’ problems. In essence, it relies upon decomposing the decision vector into subvectors, updating sequentially by minimizing an augmented Lagrangian function, and eventually updating the Lagrange multiplier. The ADMM algorithm has typically been used for solving problems in which the decision variable is decomposed into two or multiple subvec- tors. We demonstrate six template algorithms and their applica- tions and measure a peak energy efﬁciency of 279 GOPS/W.